Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

ASUS ROG G701VI - GTX1080- RED 8K Test

P

prakash d

Well-known member
Joined
Dec 10, 2014
Messages
46
Reaction score
0
Points
0
Hello Folks,


Today I got access to ASUS ROG G701VI laptop with 64 GB RAM and GTX 1080 GPU.

I tested RED 8K, 6k and 4k Sample footage from RED.COM on Premiere Pro CC 2017.

The playback in full resolution dropped 600 frames.

And when I checked the GPU Load in GPU-Z Tool only an average of 25% is used. CPU is used 100%.

I was expecting GTX 1080 would run smoothly.

Can anyone explain this?

  • Why such low GPU usage?
  • Is Premiere Pro not using full GPU Cores?
  • Is there a way to make Premiere Pro use maximum GPU?
 
Last edited:
Hey there, CUDA is installed and Premiere is using the CUDA engine for playback? To be honest, I'm still not sure the 1080 will run 8K material at full resolution without any hiccups. Though it is interesting that only a portion of your GPU is being used. It may be a system thing as well - resources dedicated.

Mostly, I'm also not very sure how the 8K support is, in being so fresh. Have you tried some 6K Dragon footage to test?
 
Tested again. The Premiere Pro Project settings has CUDA enabled.

8k -- dropped 650 frames
6k -- dropped 200 frames
5k -- dropped 70 frames
4k -- dropped 15 frames

Through out these tests GPU is only utilized an average of 25%. It did hit 40% for just one second.

But CPU is fully utilized.
 
prakash d said:
Hello Folks,


Today I got access to ASUS ROG G701VI laptop with 64 GB RAM and GTX 1080 GPU.

I tested RED 8K, 6k and 4k Sample footage from RED.COM on Premiere Pro CC 2017.

The playback in full resolution dropped 600 frames.

And when I checked the GPU Load in GPU-Z Tool only an average of 25% is used. CPU is used 100%.

I was expecting GTX 1080 would run smoothly.

Can anyone explain this?

  • Why such low GPU usage?
  • Is Premiere Pro not using full GPU Cores?
  • Is there a way to make Premiere Pro use maximum GPU?
Click to expand...

The answer or clue is right in front of you. The decode data has tapped out the CPU(s). The GPU is idle waiting on decoded data. When the CPU utilization lowers, it will provide you a better chance of testing the capabilities of the video card. The general consensus is that running 1/2 res resolution or below is the sweet spot for mid-range systems.
 
Thanks everyone for the replies.

I wonder how some folks are doing benchmarking with GTX cards and measuring which GPU is top.
 
Last edited:
I use a GTX1080 8GB alongside 32GB ram and a Quad core 4.4GHz processor. For reference, my setup can play 6K dragon footage at 1/4 (equivalent to about 5.3 megapixels worth of resolution = half of 4K but twice the res of 2K) without any problem which I find is adequate for me.
 
Last edited:
GPU Performance

GPU Performance

prakash d said:
Thanks everyone for the replies.

I wonder how some folks are doing benchmarking with GTX cards and measuring which GPU is top.
Click to expand...

The synthetic benchmark tools are most probably most useful for building 'apple to apples' profiles. These simulations generally are structured or weighted to exposed or highlight key stress areas of the hardware. So there is value in that feedback matrix. What doesn't translate well is real world performance. That's where application benchmarks serve this purpose better. But the caveat being average consumer systems are too varied to get good baseline data. Without a good control environment, the feedback data is corrupt. Often in the real world, an aging workstation system will be 'juiced up' by installing a high powered GPU.

But specific to the GPU conversation. Gaming by in large is driving the GPU market, where FPS is important. Vendors will often tune their products to have glowing industry-standard benchmark reviews. Thus we enter uncanny valley, where the user real world experience may be substantially less than the benchmarks may suggest. Particularly as we attempt to wedge that tech into the post world.

There are various ways a GPU can loose it's sexiness. Caveats being:

1) Unbalanced system. CPU, GPU, RAM and Disk I/O should have plenty of elbow. There should be very little redlining happening outside of a strenuous render session.

2) Be aware that in the search for CUDA nirvana - GPU operating frequency AND CUDA cores matter in the successful computational equation.

3) GPU optimization. Read the application's fine print on what the GPU is allowed to touch. Sometimes, it's just a few accelerated affects. Can the application benefit from multiple GPU's? Occasionally there will be vendor omissions regarding glaring application GPU inefficiencies or non optimizations.

4) Moderate user expectation: Don't insist on demanding the CPU/GPU work at the extreme levels of computation deprivation when another accommodating route can be taken for simple stuff. Basic editorial tasks can be managed by displaying lower resolution, punchy, fluid - realtime feedback is key for a good user or client edit session.

NLE performance will continue to degrade factoring in mixed media codecs, effects & nodes, sequence length, etc.

Overall, the industry is in a good space with computer hardware. $10-$15K is probably the upper coin/performance limit of non-proprietary computer hardware. In general, the hardware should not be getting in the project way of a power user. If so, workflow would be a good review point to iron wrinkles and institute some best practices.

"You can have it all. Just not all at once."
 
Chad Lancaster said:
Also a single GTX 1080 isn't going to give you real time at 8k full debayer. Some heavy lifters use resolve on supermicro GPU servers with dual high end Xeon CPU and several Titan-X Pascals
Click to expand...

No WAY..for now with 8k Helium full debayer in Davinci.
my Supermicro with dual 2699v3 and 2 1080 get 6kFF 5:1 redcode full debayer in realtime 25 fps, 8kFF 5:1 full debayer i have 18 fps.
Only solution is Redrocket-x. Or use render cache.
 
What would be the practical benefit of 8K raw realtime playback in 2016 ?
 
Yep, CPU limited. But things are usually not so simple. Even if GPU can't complete wavelet decoding/coding efficiently, you maybe able to complete part of it on the GPU, speeding up the decode.

Another possibility, is to transform the data from one domain/form to another the processing unit is good at. An example was converting from floating point domain to integer. Still, if you guys got together you could approach Nvidua about minimal instruction set/circuit to enable CUDA to be able to do it on card. This may even turn out to be one or two new instructions. It is worth trying (and talking to Open CL, Microsoft, Apple and AMD) as this is a major industry workflow use, and wavelets would he used wider. It is simpler than a comojete hardware encoder/decoder, and nor as many licensing hassles to Nvidus, as anybody that makes an actual decoder would have to get a license (as long as the minimal instructions don't violate patents). However, in the real world 2d-5d wavelets ate used for security footage that a simple Nvidia arm chipset could offer instead of competing chiosets. So, there is some usefulness there (and I wonder what 5D wavelet performance is like compared to h265 on highly compressed security applications).
 
Hrvoje. Appart from whatever cinema or 8k TV (or whatever ambarella device comes out first with it. In the old days,1080 ambarella pocket cameras were out (I had an old 16 stop 720p one due to it doing 60p, then fullhd) when people were asking about what could you do with the new fullhd TV's, and people were) those 180/360 degree cameras. They use 4k for an 180/360 degrees, which is rather pathetic, as you want 4k-8k per 60 degrees (and I haven't done the calculation of the extra resolution you need to avoid targeted resolution dewarping resolution loss).
 
The reason for this is #1 you're CPU bottlenecked. And #2, you're drive speed is probably too slow. Even SSDs are too slow at this point lol...
 
Can you get around the drive problem by maximising main memory, to cache/buffer evetything Clark? Is it a matter of the drives not keeping up with a 800MB/s+ work flow at peak highest modes, or more working memory caching?

Actually, this is a good question, people buy flash drivrs to turbo their machines, but can memory replace this?

In the old days they had battery backed ram drive PCI x cards (no good if by be restriction is actually datarate footage is brought in and out at. However, things like Intel's flash replacement and resistive ram (look it up) that acts like normal memory but retains memory like storage after power goes off would be good on this for situatikbs where its increasing the footage in and out of memory that is the issue). Can't we just have what was the name of that redray codec. Be easy to stream the footage in and out of memory then.
 
You must log in or register to reply here.
Back
Top